Denoise · Twitter

The conversation is shifting from simple prompt injection to stateful, cross-tool agent exploits, with @AnthropicAI and @GoogleDeepMind leading public research.

The focus is on red-teaming and securing autonomous agents, with major labs releasing frameworks and disclosures on complex vulnerabilities.

• Anthropic@AnthropicAIrising

  Responsible disclosure on a Claude jailbreak chain we patched last week. Full write-up including our red team timeline.

• Google DeepMind@GoogleDeepMindrising

  New red team framework for prompt injection in autonomous agents. Covers cross-tool leakage, scanner evasion, and sandbox escape patterns.

• MalwareTech@MalwareTechBlogrepeated

  Autonomous agent running pentest flows against a real SaaS. First real-world run: fewer false positives than I expected on the vulnerability surface.

Competition is moving up the stack from APIs to developer tools, with @AnthropicAI's agent challenging the IDE-centric Copilot model and getting early praise from developers like @levelsio.

A major shift towards terminal-native coding agents is underway, led by Anthropic's Claude Code 1.5 release and validated by community adoption.

• Anthropic@AnthropicAIrising

  Claude Code 1.5 is live. Terminal-native coding agent with full Claude Opus reasoning, file-ops sandbox, and session replay.

• Andrej Karpathy@karpathyrising

  The developer-experience shift from IDE to terminal agent is underrated. Coding workflows are about to look nothing like 2024.

• swyx@swyxrising

  Codex vs Claude Code terminal agent benchmarks. Pass@1 diverges more than I expected on the long-context editor tasks.

• DSPy@dspy_airising

  DSPy 3.0: prompt optimization via compile-time search over system prompt variations. Benchmarks inside.

• @levelsio@levelsiorising

  Switched my whole editor setup to Claude Code this week. Shipping faster than when I used Cursor + Copilot.

A de-facto agent stack is emerging, with protocols from @OpenAI, implementation guides from @LangChainAI, and deployment targets from @vercel and @replit converging.

Infrastructure providers are racing to support the deployment and orchestration of multi-step AI agents with new SDKs and hosting environments.

• OpenAI@OpenAIrising

  New agent SDK: protocol-level tool calling, deployment harness, and multi-worker orchestration primitives. Docs live.

• LangChain@LangChainAIrising

  MCP protocol integration thread. How to wire existing LangGraph agents into the Anthropic Model Context Protocol server spec.

• Vercel@vercelrising

  Edge runtime for agent workers is live. Spawn durable background agents from any serverless deployment.

• Alex Albert@AlexAlbert__rising

  When your security scanner finds nothing scary on an agent deploy, check the orchestration layer again. That's usually where the jailbreak sneaks through.

• Replit@replitrising

  New agent deployment harness. One command to go from local orchestration to hosted agent worker.

While agents dominate today's discourse, @MistralAI's data release shows a continued focus on improving foundational multimodal capabilities through open resources.

Mistral AI contributed a massive open dataset for web OCR, fueling development in multimodal document understanding.

• Mistral AI@MistralAIrising

  Open dataset release: 100M-row web OCR dataset. Cleaned, licensed, ready to train.

A convergence is visible between @GregKamradt's 'context engineering' concept and new layered memory systems from startups like @mem0ai, suggesting vector search alone is now considered insufficient.

The discourse is moving beyond simple RAG towards sophisticated 'context engineering' and structured agent memory systems.

• Vaibhav Srivastav@reach_vbrising

  Tested the new 10M context memory window end to end. Surprising failure modes around rag retrieval cache invalidation, thread below.

• Greg Kamradt@GregKamradtrising

  RAG is dead, long live context engineering. My framework for when to cache, when to retrieve, and when to just dump memory into the prompt.

• mem0@mem0airising

  Memory layer for agents: differentiating working memory from the subconscious store. Vector index isn't enough anymore.

• LlamaIndex@llamaindexrepeated

  Knowledge graph retrieval walkthrough: when semantic vector search misses, graph hop beats it every time.

While AI-native companies build explicit agents, established SaaS players like @NotionHQ and @linear are embedding autonomous features into existing workflows, suggesting a broader market convergence.

Workspace automation tools are shipping agent-like capabilities for tasks like issue triage and database updates.

• Notion@NotionHQrising

  Notion workspace automation is out of beta. Auto-fill tables, chained updates across databases, and a new audit log surface.

• Linear@linearrising

  Linear now auto-triages incoming issues. Quiet launch, but already our favorite workspace feature of the year.

• Temporal@temporaliorepeated

  Orchestrating agents with durable workflows: replayable, resumable, and multi-worker by default. Walkthrough from our infra team.

• James Clear@jamesclearrepeated

  The best habit tracker is the one you actually open. Three open-source alternatives worth trying.

The 'art' of prompt engineering is being industrialized into a science by platforms like @weights_biases, enabling quantitative analysis that moves beyond anecdotal tricks.

The focus is on systematic, large-scale benchmarking of system prompts to find an efficient frontier for production models.

• dotey@doteyrising

  Five prompt tricks learned this week from reviewing 200 production prompts. Short thread.

• Weights & Biases@weights_biasesrising

  System prompt benchmarking at scale: we ran 40k variants across 6 frontier models. The efficient frontier is not where you think.

The bottleneck in agent development appears to be shifting from raw compute to sophisticated data filtering, as noted by @jerryjliu0's work on preventing data poisoning.

A key challenge highlighted is the curation of high-quality synthetic data for training robust and generalizable agents.

• Jerry Liu@jerryjliu0repeated

  Dataset curation for agent training: how we filter synthetic data that looks good but poisons generalization.